Realtime Visual Recognition in deep

Convolutional Neural Networks

Objective of the Project
In recent days the detection of visual projects is done with convolutional
neural networks(CNN).Later the development of fully convolutional neural
networks(FCN) ,moreover there are chances for improvisation over generic FCN
models which deals with scale space problems. Also on the other side
holistically-nested edge detector provides a skip layer structure with deep
supervision for edge and boundry detection.
Our framework takes full advantage of extracting from FCN providing more
advanced representation at each layer,this property is used for segment detection.
It has advantages in terms of efficiency ie ,0.08 second per
image,effectiveness and simplicity over existing algorithms.
It can be analysed on the role of training data on performance, the
experimental results provide a more reasonable and powerful training set for
future research.
Introduction
Human beings have the ability of identifying the region or object
in an image. Visual Recognition aims to detecting and identifying
the object or region in an image automatically.Earlier methods for
salient detection are mainly inspired by cognitive studies of visual
attention.
Our framework detects all the objects based on training set
provided to it. Main view of this project is to increase the
accuracy rate of the detection ie, it recoginze the object even in
blur stage or under less brightness
Over 60-65 images can be trained for detection
This framework can be constructed using deep learning , which is
a subset of machine learning in AI that has network capability of
learing unsupervised from data that is unstructured or unabled
 Project Feasibility Study
By using deep learing this framework can be implemented
, which recoginze and detects the objects effectively.
Python also plays a major role in executing the framework

Over the past two decades the saliency detection methods

like CNN based saliency models were used which cant
detect the blur objects.Here we mainly focus on using
recent salient object detection method based on deep
learning architectures

Anaconda , python are the technologies used.

SYSTEM CONFIGURATION
HardWare SYSTEM CONFIGURATION
 Processor - Intel Core i7
 RAM - 4 GB
 Hard Disk - 500 GB

SoftWare SYSTEM CONFIGURATION

 Operating System - Windows 7/8/10
 Programming Language : Python 3.7
 Anaconda
Proposed system
Deep CNNs have been extensively used for object detection.
 CNN is a type of feed-forward neural network and works on
principle of weight sharing. Convolution is an integration showing
how one function overlaps with other function and is a blend of
two functions being multiplied.
Image is convolved with activation function to get feature maps.
To reduce spatial complexity of the network, feature maps are
treated with pooling layers to get abstracted feature maps. This
process is repeated for the desired number of filters and
accordingly feature maps are created. Eventually, these feature
maps are processed with fully connected layers to get output of
image recognition showing confidence score for the predicted
class labels.
Applications
Object detection is applicable in many domains
ranging from defense (surveillance), human computer
interaction, robotics, transportation, retrieval, etc.
Sensors used for persistent surveillance generate
petabytes of image data in few hours.
Modules
 Object Data collection.
Data Preprocessing.
Yolo algorithm.
Object Data collection.
Real time data collected from kaggle .Collection of
data is one of the major and most important tasks of
any machine learning projects. Because the input we
feed to the algorithms is data. So, the algorithms
efficiency and accuracy depends upon the correctness
and quality of data collected. So as the data same will
be the output.
Data Preprocessing.

Collecting the data is one task and making that data useful is an-other
vital task.
Data collected from various means will be in an unorganized format and
there may be lot of null values, in-valid data values and unwanted data.
Cleaning all these data and replacing them with appropriate or
approximate data and removing null and missing data and replacing
them with some fixed alternate values are the basic steps in pre
processing of data.
Even data collected may contain completely garbage values. It may not
be in exact format or way that is meant to be. All such cases must be
verified and replaced with alternate values to make data meaning
meaningful and useful for further processing. Data must be kept in a
organized format.
Yolo Algorithm
Yolo is an algorithm that uses convolutional neural
networks for object detection.
In comparison to recognition algorithms, a detection
algorithm does not only predict class labels, but detects
locations of objects as well.
The algorithm divides the image into grids and runs the
image classification and localization algorithm
(discussed under object localization) on each of the grid
cells. For example, we have an input image of size 256 ×
256. We place a 3 × 3 grid on the image .
References
1.P. F. Felzenszwalb, R. B. Girshick, D. McAllester, and D. Ramanan, “Object detection with
discriminatively trained part-based models,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 32, no. 9, pp.
1627 1645, Sep. 2010.
2. B. Leibe, A. Leonardis, and B. Schiele, “Robust object detection with interleaved categorization and
segmentation,” Int. J. Comput. Vis., vol. 77, nos. 1-3, pp. 259-289, May 2008.
3. J. Zhang, M. Marszalek, S. Lazebnik, and C. Schmid, “Local features and kernels for classi cation of
texture and object categories: A comprehensive study,” in Proc. Conf. Comput. Vis. Pattern
Recognit.Workshop (CVPRW), Jun. 2006, p. 13.
4. P. Viola and M. Jones, “Rapid object detection using a boosted cascade of simple features,” in Proc.
IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit. (CVPR), vol. 1. Dec. 2001, pp. 511-518.
5. M. Weber, M. Welling, and P. Perona, “Towards automatic discovery of object categories,” in Proc.
IEEE Conf. Comput. Vis. Pattern Recognit., vol. 2. Jun. 2000, pp. 101-108.
6. A. Ayvaci and S. Soatto, “Detachable object detection: Segmentation and depth ordering from short-
baseline video,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 34, no. 10, pp. 1942 1951, Oct. 2012.
7. D. Liu, M.-L. Shyu, Q. Zhu, and S.-C. Chen, “Moving object detection under object occlusion
situations in video sequences,'' in Proc. IEEE Int. Symp. Multimedia (ISM), Dec. 2011, pp. 271-278.
8. J. Kim, G. Ye, and D. Kim, ``Moving object detection under free-moving camera,'' in Proc. 17th IEEE
Int. Conf. Image Process. (ICIP), Sep. 2010, pp. 4669-4672.
9.B. Qi, M. Ghazal, and A. Amer, ``Robust global
motion estimation oriented to video object
segmentation,'' IEEE Trans. Image Process., vol. 17,
no. 6, pp. 958 967, Jun. 2008.
10. S. Kumar and M. Hebert, ``A hierarchical eld
framework for unified context-based classi cation,'' in
Proc. 10th IEEE Int. Conf. Comput. Vis. (ICCV), vol.
2. Oct. 2005, pp. 1284-1291.